kernel/trace/blktrace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4  *
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/blktrace_api.h>
  12 #include <linux/percpu.h>
  13 #include <linux/init.h>
  14 #include <linux/mutex.h>
  15 #include <linux/slab.h>
  16 #include <linux/debugfs.h>
  17 #include <linux/export.h>
  18 #include <linux/time.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/list.h>
  21 #include <linux/blk-cgroup.h>
  22
  23 #include "../../block/blk.h"
  24
  25 #include <trace/events/block.h>
  26
  27 #include "trace_output.h"
  28
  29 #ifdef CONFIG_BLK_DEV_IO_TRACE
  30
  31 static unsigned int blktrace_seq __read_mostly = 1;
  32
  33 static struct trace_array *blk_tr;
  34 static bool blk_tracer_enabled __read_mostly;
  35
  36 static LIST_HEAD(running_trace_list);
  37 static __cacheline_aligned_in_smp DEFINE_RAW_SPINLOCK(running_trace_lock);
  38
  39 /* Select an alternative, minimalistic output than the original one */
  40 #define TRACE_BLK_OPT_CLASSIC   0x1
  41 #define TRACE_BLK_OPT_CGROUP    0x2
  42 #define TRACE_BLK_OPT_CGNAME    0x4
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47 #ifdef CONFIG_BLK_CGROUP
  48         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  49         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  50 #endif
  51         { }
  52 };
  53
  54 static struct tracer_flags blk_tracer_flags = {
  55         .val  = 0,
  56         .opts = blk_tracer_opts,
  57 };
  58
  59 /* Global reference count of probes */
  60 static DEFINE_MUTEX(blk_probe_mutex);
  61 static int blk_probes_ref;
  62
  63 static void blk_register_tracepoints(void);
  64 static void blk_unregister_tracepoints(void);
  65
  66 /*
  67  * Send out a notify message.
  68  */
  69 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  70                        const void *data, size_t len, u64 cgid)
  71 {
  72         struct blk_io_trace *t;
  73         struct ring_buffer_event *event = NULL;
  74         struct trace_buffer *buffer = NULL;
  75         unsigned int trace_ctx = 0;
  76         int cpu = smp_processor_id();
  77         bool blk_tracer = blk_tracer_enabled;
  78         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
  79
  80         if (blk_tracer) {
  81                 buffer = blk_tr->array_buffer.buffer;
  82                 trace_ctx = tracing_gen_ctx_flags(0);
  83                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84                                                   sizeof(*t) + len + cgid_len,
  85                                                   trace_ctx);
  86                 if (!event)
  87                         return;
  88                 t = ring_buffer_event_data(event);
  89                 goto record_it;
  90         }
  91
  92         if (!bt->rchan)
  93                 return;
  94
  95         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  96         if (t) {
  97                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  98                 t->time = ktime_to_ns(ktime_get());
  99 record_it:
 100                 t->device = bt->dev;
 101                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 102                 t->pid = pid;
 103                 t->cpu = cpu;
 104                 t->pdu_len = len + cgid_len;
 105                 if (cgid_len)
 106                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 107                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 108
 109                 if (blk_tracer)
 110                         trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx);
 111         }
 112 }
 113
 114 /*
 115  * Send out a notify for this process, if we haven't done so since a trace
 116  * started
 117  */
 118 static void trace_note_tsk(struct task_struct *tsk)
 119 {
 120         unsigned long flags;
 121         struct blk_trace *bt;
 122
 123         tsk->btrace_seq = blktrace_seq;
 124         raw_spin_lock_irqsave(&running_trace_lock, flags);
 125         list_for_each_entry(bt, &running_trace_list, running_list) {
 126                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 127                            sizeof(tsk->comm), 0);
 128         }
 129         raw_spin_unlock_irqrestore(&running_trace_lock, flags);
 130 }
 131
 132 static void trace_note_time(struct blk_trace *bt)
 133 {
 134         struct timespec64 now;
 135         unsigned long flags;
 136         u32 words[2];
 137
 138         /* need to check user space to see if this breaks in y2038 or y2106 */
 139         ktime_get_real_ts64(&now);
 140         words[0] = (u32)now.tv_sec;
 141         words[1] = now.tv_nsec;
 142
 143         local_irq_save(flags);
 144         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
 145         local_irq_restore(flags);
 146 }
 147
 148 void __blk_trace_note_message(struct blk_trace *bt,
 149                 struct cgroup_subsys_state *css, const char *fmt, ...)
 150 {
 151         int n;
 152         va_list args;
 153         unsigned long flags;
 154         char *buf;
 155         u64 cgid = 0;
 156
 157         if (unlikely(bt->trace_state != Blktrace_running &&
 158                      !blk_tracer_enabled))
 159                 return;
 160
 161         /*
 162          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 163          * message to the trace.
 164          */
 165         if (!(bt->act_mask & BLK_TC_NOTIFY))
 166                 return;
 167
 168         local_irq_save(flags);
 169         buf = this_cpu_ptr(bt->msg_data);
 170         va_start(args, fmt);
 171         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 172         va_end(args);
 173
 174 #ifdef CONFIG_BLK_CGROUP
 175         if (css && (blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 176                 cgid = cgroup_id(css->cgroup);
 177         else
 178                 cgid = 1;
 179 #endif
 180         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, cgid);
 181         local_irq_restore(flags);
 182 }
 183 EXPORT_SYMBOL_GPL(__blk_trace_note_message);
 184
 185 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 186                          pid_t pid)
 187 {
 188         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 189                 return 1;
 190         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 191                 return 1;
 192         if (bt->pid && pid != bt->pid)
 193                 return 1;
 194
 195         return 0;
 196 }
 197
 198 /*
 199  * Data direction bit lookup
 200  */
 201 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 202                                  BLK_TC_ACT(BLK_TC_WRITE) };
 203
 204 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 205 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 206
 207 /* The ilog2() calls fall out because they're constant */
 208 #define MASK_TC_BIT(rw, __name) ((__force u32)(rw & REQ_ ## __name) <<  \
 209           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 210
 211 /*
 212  * The worker for the various blk_add_trace*() types. Fills out a
 213  * blk_io_trace structure and places it in a per-cpu subbuffer.
 214  */
 215 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 216                             const blk_opf_t opf, u32 what, int error,
 217                             int pdu_len, void *pdu_data, u64 cgid)
 218 {
 219         struct task_struct *tsk = current;
 220         struct ring_buffer_event *event = NULL;
 221         struct trace_buffer *buffer = NULL;
 222         struct blk_io_trace *t;
 223         unsigned long flags = 0;
 224         unsigned long *sequence;
 225         unsigned int trace_ctx = 0;
 226         pid_t pid;
 227         int cpu;
 228         bool blk_tracer = blk_tracer_enabled;
 229         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
 230         const enum req_op op = opf & REQ_OP_MASK;
 231
 232         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 233                 return;
 234
 235         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 236         what |= MASK_TC_BIT(opf, SYNC);
 237         what |= MASK_TC_BIT(opf, RAHEAD);
 238         what |= MASK_TC_BIT(opf, META);
 239         what |= MASK_TC_BIT(opf, PREFLUSH);
 240         what |= MASK_TC_BIT(opf, FUA);
 241         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 242                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 243         if (op == REQ_OP_FLUSH)
 244                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 245         if (cgid)
 246                 what |= __BLK_TA_CGROUP;
 247
 248         pid = tsk->pid;
 249         if (act_log_check(bt, what, sector, pid))
 250                 return;
 251         cpu = raw_smp_processor_id();
 252
 253         if (blk_tracer) {
 254                 tracing_record_cmdline(current);
 255
 256                 buffer = blk_tr->array_buffer.buffer;
 257                 trace_ctx = tracing_gen_ctx_flags(0);
 258                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 259                                                   sizeof(*t) + pdu_len + cgid_len,
 260                                                   trace_ctx);
 261                 if (!event)
 262                         return;
 263                 t = ring_buffer_event_data(event);
 264                 goto record_it;
 265         }
 266
 267         if (unlikely(tsk->btrace_seq != blktrace_seq))
 268                 trace_note_tsk(tsk);
 269
 270         /*
 271          * A word about the locking here - we disable interrupts to reserve
 272          * some space in the relay per-cpu buffer, to prevent an irq
 273          * from coming in and stepping on our toes.
 274          */
 275         local_irq_save(flags);
 276         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 277         if (t) {
 278                 sequence = per_cpu_ptr(bt->sequence, cpu);
 279
 280                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 281                 t->sequence = ++(*sequence);
 282                 t->time = ktime_to_ns(ktime_get());
 283 record_it:
 284                 /*
 285                  * These two are not needed in ftrace as they are in the
 286                  * generic trace_entry, filled by tracing_generic_entry_update,
 287                  * but for the trace_event->bin() synthesizer benefit we do it
 288                  * here too.
 289                  */
 290                 t->cpu = cpu;
 291                 t->pid = pid;
 292
 293                 t->sector = sector;
 294                 t->bytes = bytes;
 295                 t->action = what;
 296                 t->device = bt->dev;
 297                 t->error = error;
 298                 t->pdu_len = pdu_len + cgid_len;
 299
 300                 if (cgid_len)
 301                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 302                 if (pdu_len)
 303                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 304
 305                 if (blk_tracer) {
 306                         trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx);
 307                         return;
 308                 }
 309         }
 310
 311         local_irq_restore(flags);
 312 }
 313
 314 static void blk_trace_free(struct request_queue *q, struct blk_trace *bt)
 315 {
 316         relay_close(bt->rchan);
 317
 318         /*
 319          * If 'bt->dir' is not set, then both 'dropped' and 'msg' are created
 320          * under 'q->debugfs_dir', thus lookup and remove them.
 321          */
 322         if (!bt->dir) {
 323                 debugfs_lookup_and_remove("dropped", q->debugfs_dir);
 324                 debugfs_lookup_and_remove("msg", q->debugfs_dir);
 325         } else {
 326                 debugfs_remove(bt->dir);
 327         }
 328         free_percpu(bt->sequence);
 329         free_percpu(bt->msg_data);
 330         kfree(bt);
 331 }
 332
 333 static void get_probe_ref(void)
 334 {
 335         mutex_lock(&blk_probe_mutex);
 336         if (++blk_probes_ref == 1)
 337                 blk_register_tracepoints();
 338         mutex_unlock(&blk_probe_mutex);
 339 }
 340
 341 static void put_probe_ref(void)
 342 {
 343         mutex_lock(&blk_probe_mutex);
 344         if (!--blk_probes_ref)
 345                 blk_unregister_tracepoints();
 346         mutex_unlock(&blk_probe_mutex);
 347 }
 348
 349 static int blk_trace_start(struct blk_trace *bt)
 350 {
 351         if (bt->trace_state != Blktrace_setup &&
 352             bt->trace_state != Blktrace_stopped)
 353                 return -EINVAL;
 354
 355         blktrace_seq++;
 356         smp_mb();
 357         bt->trace_state = Blktrace_running;
 358         raw_spin_lock_irq(&running_trace_lock);
 359         list_add(&bt->running_list, &running_trace_list);
 360         raw_spin_unlock_irq(&running_trace_lock);
 361         trace_note_time(bt);
 362
 363         return 0;
 364 }
 365
 366 static int blk_trace_stop(struct blk_trace *bt)
 367 {
 368         if (bt->trace_state != Blktrace_running)
 369                 return -EINVAL;
 370
 371         bt->trace_state = Blktrace_stopped;
 372         raw_spin_lock_irq(&running_trace_lock);
 373         list_del_init(&bt->running_list);
 374         raw_spin_unlock_irq(&running_trace_lock);
 375         relay_flush(bt->rchan);
 376
 377         return 0;
 378 }
 379
 380 static void blk_trace_cleanup(struct request_queue *q, struct blk_trace *bt)
 381 {
 382         blk_trace_stop(bt);
 383         synchronize_rcu();
 384         blk_trace_free(q, bt);
 385         put_probe_ref();
 386 }
 387
 388 static int __blk_trace_remove(struct request_queue *q)
 389 {
 390         struct blk_trace *bt;
 391
 392         bt = rcu_replace_pointer(q->blk_trace, NULL,
 393                                  lockdep_is_held(&q->debugfs_mutex));
 394         if (!bt)
 395                 return -EINVAL;
 396
 397         blk_trace_cleanup(q, bt);
 398
 399         return 0;
 400 }
 401
 402 int blk_trace_remove(struct request_queue *q)
 403 {
 404         int ret;
 405
 406         mutex_lock(&q->debugfs_mutex);
 407         ret = __blk_trace_remove(q);
 408         mutex_unlock(&q->debugfs_mutex);
 409
 410         return ret;
 411 }
 412 EXPORT_SYMBOL_GPL(blk_trace_remove);
 413
 414 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 415                                 size_t count, loff_t *ppos)
 416 {
 417         struct blk_trace *bt = filp->private_data;
 418         char buf[16];
 419
 420         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 421
 422         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 423 }
 424
 425 static const struct file_operations blk_dropped_fops = {
 426         .owner =        THIS_MODULE,
 427         .open =         simple_open,
 428         .read =         blk_dropped_read,
 429         .llseek =       default_llseek,
 430 };
 431
 432 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 433                                 size_t count, loff_t *ppos)
 434 {
 435         char *msg;
 436         struct blk_trace *bt;
 437
 438         if (count >= BLK_TN_MAX_MSG)
 439                 return -EINVAL;
 440
 441         msg = memdup_user_nul(buffer, count);
 442         if (IS_ERR(msg))
 443                 return PTR_ERR(msg);
 444
 445         bt = filp->private_data;
 446         __blk_trace_note_message(bt, NULL, "%s", msg);
 447         kfree(msg);
 448
 449         return count;
 450 }
 451
 452 static const struct file_operations blk_msg_fops = {
 453         .owner =        THIS_MODULE,
 454         .open =         simple_open,
 455         .write =        blk_msg_write,
 456         .llseek =       noop_llseek,
 457 };
 458
 459 /*
 460  * Keep track of how many times we encountered a full subbuffer, to aid
 461  * the user space app in telling how many lost events there were.
 462  */
 463 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 464                                      void *prev_subbuf, size_t prev_padding)
 465 {
 466         struct blk_trace *bt;
 467
 468         if (!relay_buf_full(buf))
 469                 return 1;
 470
 471         bt = buf->chan->private_data;
 472         atomic_inc(&bt->dropped);
 473         return 0;
 474 }
 475
 476 static int blk_remove_buf_file_callback(struct dentry *dentry)
 477 {
 478         debugfs_remove(dentry);
 479
 480         return 0;
 481 }
 482
 483 static struct dentry *blk_create_buf_file_callback(const char *filename,
 484                                                    struct dentry *parent,
 485                                                    umode_t mode,
 486                                                    struct rchan_buf *buf,
 487                                                    int *is_global)
 488 {
 489         return debugfs_create_file(filename, mode, parent, buf,
 490                                         &relay_file_operations);
 491 }
 492
 493 static const struct rchan_callbacks blk_relay_callbacks = {
 494         .subbuf_start           = blk_subbuf_start_callback,
 495         .create_buf_file        = blk_create_buf_file_callback,
 496         .remove_buf_file        = blk_remove_buf_file_callback,
 497 };
 498
 499 static void blk_trace_setup_lba(struct blk_trace *bt,
 500                                 struct block_device *bdev)
 501 {
 502         if (bdev) {
 503                 bt->start_lba = bdev->bd_start_sect;
 504                 bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
 505         } else {
 506                 bt->start_lba = 0;
 507                 bt->end_lba = -1ULL;
 508         }
 509 }
 510
 511 /*
 512  * Setup everything required to start tracing
 513  */
 514 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 515                               struct block_device *bdev,
 516                               struct blk_user_trace_setup *buts)
 517 {
 518         struct blk_trace *bt = NULL;
 519         struct dentry *dir = NULL;
 520         int ret;
 521
 522         lockdep_assert_held(&q->debugfs_mutex);
 523
 524         if (!buts->buf_size || !buts->buf_nr)
 525                 return -EINVAL;
 526
 527         strscpy_pad(buts->name, name, BLKTRACE_BDEV_SIZE);
 528
 529         /*
 530          * some device names have larger paths - convert the slashes
 531          * to underscores for this to work as expected
 532          */
 533         strreplace(buts->name, '/', '_');
 534
 535         /*
 536          * bdev can be NULL, as with scsi-generic, this is a helpful as
 537          * we can be.
 538          */
 539         if (rcu_dereference_protected(q->blk_trace,
 540                                       lockdep_is_held(&q->debugfs_mutex))) {
 541                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 542                         buts->name);
 543                 return -EBUSY;
 544         }
 545
 546         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 547         if (!bt)
 548                 return -ENOMEM;
 549
 550         ret = -ENOMEM;
 551         bt->sequence = alloc_percpu(unsigned long);
 552         if (!bt->sequence)
 553                 goto err;
 554
 555         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 556         if (!bt->msg_data)
 557                 goto err;
 558
 559         /*
 560          * When tracing the whole disk reuse the existing debugfs directory
 561          * created by the block layer on init. For partitions block devices,
 562          * and scsi-generic block devices we create a temporary new debugfs
 563          * directory that will be removed once the trace ends.
 564          */
 565         if (bdev && !bdev_is_partition(bdev))
 566                 dir = q->debugfs_dir;
 567         else
 568                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 569
 570         /*
 571          * As blktrace relies on debugfs for its interface the debugfs directory
 572          * is required, contrary to the usual mantra of not checking for debugfs
 573          * files or directories.
 574          */
 575         if (IS_ERR_OR_NULL(dir)) {
 576                 pr_warn("debugfs_dir not present for %s so skipping\n",
 577                         buts->name);
 578                 ret = -ENOENT;
 579                 goto err;
 580         }
 581
 582         bt->dev = dev;
 583         atomic_set(&bt->dropped, 0);
 584         INIT_LIST_HEAD(&bt->running_list);
 585
 586         ret = -EIO;
 587         debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 588         debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 589
 590         bt->rchan = relay_open("trace", dir, buts->buf_size,
 591                                 buts->buf_nr, &blk_relay_callbacks, bt);
 592         if (!bt->rchan)
 593                 goto err;
 594
 595         bt->act_mask = buts->act_mask;
 596         if (!bt->act_mask)
 597                 bt->act_mask = (u16) -1;
 598
 599         blk_trace_setup_lba(bt, bdev);
 600
 601         /* overwrite with user settings */
 602         if (buts->start_lba)
 603                 bt->start_lba = buts->start_lba;
 604         if (buts->end_lba)
 605                 bt->end_lba = buts->end_lba;
 606
 607         bt->pid = buts->pid;
 608         bt->trace_state = Blktrace_setup;
 609
 610         rcu_assign_pointer(q->blk_trace, bt);
 611         get_probe_ref();
 612
 613         ret = 0;
 614 err:
 615         if (ret)
 616                 blk_trace_free(q, bt);
 617         return ret;
 618 }
 619
 620 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 621                     struct block_device *bdev,
 622                     char __user *arg)
 623 {
 624         struct blk_user_trace_setup buts;
 625         int ret;
 626
 627         ret = copy_from_user(&buts, arg, sizeof(buts));
 628         if (ret)
 629                 return -EFAULT;
 630
 631         mutex_lock(&q->debugfs_mutex);
 632         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 633         mutex_unlock(&q->debugfs_mutex);
 634         if (ret)
 635                 return ret;
 636
 637         if (copy_to_user(arg, &buts, sizeof(buts))) {
 638                 blk_trace_remove(q);
 639                 return -EFAULT;
 640         }
 641         return 0;
 642 }
 643 EXPORT_SYMBOL_GPL(blk_trace_setup);
 644
 645 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 646 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 647                                   dev_t dev, struct block_device *bdev,
 648                                   char __user *arg)
 649 {
 650         struct blk_user_trace_setup buts;
 651         struct compat_blk_user_trace_setup cbuts;
 652         int ret;
 653
 654         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 655                 return -EFAULT;
 656
 657         buts = (struct blk_user_trace_setup) {
 658                 .act_mask = cbuts.act_mask,
 659                 .buf_size = cbuts.buf_size,
 660                 .buf_nr = cbuts.buf_nr,
 661                 .start_lba = cbuts.start_lba,
 662                 .end_lba = cbuts.end_lba,
 663                 .pid = cbuts.pid,
 664         };
 665
 666         mutex_lock(&q->debugfs_mutex);
 667         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 668         mutex_unlock(&q->debugfs_mutex);
 669         if (ret)
 670                 return ret;
 671
 672         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 673                 blk_trace_remove(q);
 674                 return -EFAULT;
 675         }
 676
 677         return 0;
 678 }
 679 #endif
 680
 681 static int __blk_trace_startstop(struct request_queue *q, int start)
 682 {
 683         struct blk_trace *bt;
 684
 685         bt = rcu_dereference_protected(q->blk_trace,
 686                                        lockdep_is_held(&q->debugfs_mutex));
 687         if (bt == NULL)
 688                 return -EINVAL;
 689
 690         if (start)
 691                 return blk_trace_start(bt);
 692         else
 693                 return blk_trace_stop(bt);
 694 }
 695
 696 int blk_trace_startstop(struct request_queue *q, int start)
 697 {
 698         int ret;
 699
 700         mutex_lock(&q->debugfs_mutex);
 701         ret = __blk_trace_startstop(q, start);
 702         mutex_unlock(&q->debugfs_mutex);
 703
 704         return ret;
 705 }
 706 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 707
 708 /*
 709  * When reading or writing the blktrace sysfs files, the references to the
 710  * opened sysfs or device files should prevent the underlying block device
 711  * from being removed. So no further delete protection is really needed.
 712  */
 713
 714 /**
 715  * blk_trace_ioctl - handle the ioctls associated with tracing
 716  * @bdev:       the block device
 717  * @cmd:        the ioctl cmd
 718  * @arg:        the argument data, if any
 719  *
 720  **/
 721 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 722 {
 723         struct request_queue *q = bdev_get_queue(bdev);
 724         int ret, start = 0;
 725         char b[BDEVNAME_SIZE];
 726
 727         switch (cmd) {
 728         case BLKTRACESETUP:
 729                 snprintf(b, sizeof(b), "%pg", bdev);
 730                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 731                 break;
 732 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 733         case BLKTRACESETUP32:
 734                 snprintf(b, sizeof(b), "%pg", bdev);
 735                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 736                 break;
 737 #endif
 738         case BLKTRACESTART:
 739                 start = 1;
 740                 fallthrough;
 741         case BLKTRACESTOP:
 742                 ret = blk_trace_startstop(q, start);
 743                 break;
 744         case BLKTRACETEARDOWN:
 745                 ret = blk_trace_remove(q);
 746                 break;
 747         default:
 748                 ret = -ENOTTY;
 749                 break;
 750         }
 751         return ret;
 752 }
 753
 754 /**
 755  * blk_trace_shutdown - stop and cleanup trace structures
 756  * @q:    the request queue associated with the device
 757  *
 758  **/
 759 void blk_trace_shutdown(struct request_queue *q)
 760 {
 761         if (rcu_dereference_protected(q->blk_trace,
 762                                       lockdep_is_held(&q->debugfs_mutex)))
 763                 __blk_trace_remove(q);
 764 }
 765
 766 #ifdef CONFIG_BLK_CGROUP
 767 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 768 {
 769         struct cgroup_subsys_state *blkcg_css;
 770         struct blk_trace *bt;
 771
 772         /* We don't use the 'bt' value here except as an optimization... */
 773         bt = rcu_dereference_protected(q->blk_trace, 1);
 774         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 775                 return 0;
 776
 777         blkcg_css = bio_blkcg_css(bio);
 778         if (!blkcg_css)
 779                 return 0;
 780         return cgroup_id(blkcg_css->cgroup);
 781 }
 782 #else
 783 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 784 {
 785         return 0;
 786 }
 787 #endif
 788
 789 static u64
 790 blk_trace_request_get_cgid(struct request *rq)
 791 {
 792         if (!rq->bio)
 793                 return 0;
 794         /* Use the first bio */
 795         return blk_trace_bio_get_cgid(rq->q, rq->bio);
 796 }
 797
 798 /*
 799  * blktrace probes
 800  */
 801
 802 /**
 803  * blk_add_trace_rq - Add a trace for a request oriented action
 804  * @rq:         the source request
 805  * @error:      return status to log
 806  * @nr_bytes:   number of completed bytes
 807  * @what:       the action
 808  * @cgid:       the cgroup info
 809  *
 810  * Description:
 811  *     Records an action against a request. Will log the bio offset + size.
 812  *
 813  **/
 814 static void blk_add_trace_rq(struct request *rq, blk_status_t error,
 815                              unsigned int nr_bytes, u32 what, u64 cgid)
 816 {
 817         struct blk_trace *bt;
 818
 819         rcu_read_lock();
 820         bt = rcu_dereference(rq->q->blk_trace);
 821         if (likely(!bt)) {
 822                 rcu_read_unlock();
 823                 return;
 824         }
 825
 826         if (blk_rq_is_passthrough(rq))
 827                 what |= BLK_TC_ACT(BLK_TC_PC);
 828         else
 829                 what |= BLK_TC_ACT(BLK_TC_FS);
 830
 831         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, rq->cmd_flags,
 832                         what, blk_status_to_errno(error), 0, NULL, cgid);
 833         rcu_read_unlock();
 834 }
 835
 836 static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
 837 {
 838         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 839                          blk_trace_request_get_cgid(rq));
 840 }
 841
 842 static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
 843 {
 844         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 845                          blk_trace_request_get_cgid(rq));
 846 }
 847
 848 static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
 849 {
 850         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
 851                          blk_trace_request_get_cgid(rq));
 852 }
 853
 854 static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
 855 {
 856         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 857                          blk_trace_request_get_cgid(rq));
 858 }
 859
 860 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 861                         blk_status_t error, unsigned int nr_bytes)
 862 {
 863         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 864                          blk_trace_request_get_cgid(rq));
 865 }
 866
 867 /**
 868  * blk_add_trace_bio - Add a trace for a bio oriented action
 869  * @q:          queue the io is for
 870  * @bio:        the source bio
 871  * @what:       the action
 872  * @error:      error, if any
 873  *
 874  * Description:
 875  *     Records an action against a bio. Will log the bio offset + size.
 876  *
 877  **/
 878 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 879                               u32 what, int error)
 880 {
 881         struct blk_trace *bt;
 882
 883         rcu_read_lock();
 884         bt = rcu_dereference(q->blk_trace);
 885         if (likely(!bt)) {
 886                 rcu_read_unlock();
 887                 return;
 888         }
 889
 890         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 891                         bio->bi_opf, what, error, 0, NULL,
 892                         blk_trace_bio_get_cgid(q, bio));
 893         rcu_read_unlock();
 894 }
 895
 896 static void blk_add_trace_bio_complete(void *ignore,
 897                                        struct request_queue *q, struct bio *bio)
 898 {
 899         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
 900                           blk_status_to_errno(bio->bi_status));
 901 }
 902
 903 static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio)
 904 {
 905         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE,
 906                         0);
 907 }
 908
 909 static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio)
 910 {
 911         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE,
 912                         0);
 913 }
 914
 915 static void blk_add_trace_bio_queue(void *ignore, struct bio *bio)
 916 {
 917         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0);
 918 }
 919
 920 static void blk_add_trace_getrq(void *ignore, struct bio *bio)
 921 {
 922         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0);
 923 }
 924
 925 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 926 {
 927         struct blk_trace *bt;
 928
 929         rcu_read_lock();
 930         bt = rcu_dereference(q->blk_trace);
 931         if (bt)
 932                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 933         rcu_read_unlock();
 934 }
 935
 936 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 937                                     unsigned int depth, bool explicit)
 938 {
 939         struct blk_trace *bt;
 940
 941         rcu_read_lock();
 942         bt = rcu_dereference(q->blk_trace);
 943         if (bt) {
 944                 __be64 rpdu = cpu_to_be64(depth);
 945                 u32 what;
 946
 947                 if (explicit)
 948                         what = BLK_TA_UNPLUG_IO;
 949                 else
 950                         what = BLK_TA_UNPLUG_TIMER;
 951
 952                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
 953         }
 954         rcu_read_unlock();
 955 }
 956
 957 static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu)
 958 {
 959         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 960         struct blk_trace *bt;
 961
 962         rcu_read_lock();
 963         bt = rcu_dereference(q->blk_trace);
 964         if (bt) {
 965                 __be64 rpdu = cpu_to_be64(pdu);
 966
 967                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 968                                 bio->bi_iter.bi_size, bio->bi_opf, BLK_TA_SPLIT,
 969                                 blk_status_to_errno(bio->bi_status),
 970                                 sizeof(rpdu), &rpdu,
 971                                 blk_trace_bio_get_cgid(q, bio));
 972         }
 973         rcu_read_unlock();
 974 }
 975
 976 /**
 977  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 978  * @ignore:     trace callback data parameter (not used)
 979  * @bio:        the source bio
 980  * @dev:        source device
 981  * @from:       source sector
 982  *
 983  * Called after a bio is remapped to a different device and/or sector.
 984  **/
 985 static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev,
 986                                     sector_t from)
 987 {
 988         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 989         struct blk_trace *bt;
 990         struct blk_io_trace_remap r;
 991
 992         rcu_read_lock();
 993         bt = rcu_dereference(q->blk_trace);
 994         if (likely(!bt)) {
 995                 rcu_read_unlock();
 996                 return;
 997         }
 998
 999         r.device_from = cpu_to_be32(dev);
1000         r.device_to   = cpu_to_be32(bio_dev(bio));
1001         r.sector_from = cpu_to_be64(from);
1002
1003         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1004                         bio->bi_opf, BLK_TA_REMAP,
1005                         blk_status_to_errno(bio->bi_status),
1006                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1007         rcu_read_unlock();
1008 }
1009
1010 /**
1011  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1012  * @ignore:     trace callback data parameter (not used)
1013  * @rq:         the source request
1014  * @dev:        target device
1015  * @from:       source sector
1016  *
1017  * Description:
1018  *     Device mapper remaps request to other devices.
1019  *     Add a trace for that action.
1020  *
1021  **/
1022 static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
1023                                    sector_t from)
1024 {
1025         struct blk_trace *bt;
1026         struct blk_io_trace_remap r;
1027
1028         rcu_read_lock();
1029         bt = rcu_dereference(rq->q->blk_trace);
1030         if (likely(!bt)) {
1031                 rcu_read_unlock();
1032                 return;
1033         }
1034
1035         r.device_from = cpu_to_be32(dev);
1036         r.device_to   = cpu_to_be32(disk_devt(rq->q->disk));
1037         r.sector_from = cpu_to_be64(from);
1038
1039         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1040                         rq->cmd_flags, BLK_TA_REMAP, 0,
1041                         sizeof(r), &r, blk_trace_request_get_cgid(rq));
1042         rcu_read_unlock();
1043 }
1044
1045 /**
1046  * blk_add_driver_data - Add binary message with driver-specific data
1047  * @rq:         io request
1048  * @data:       driver-specific data
1049  * @len:        length of driver-specific data
1050  *
1051  * Description:
1052  *     Some drivers might want to write driver-specific data per request.
1053  *
1054  **/
1055 void blk_add_driver_data(struct request *rq, void *data, size_t len)
1056 {
1057         struct blk_trace *bt;
1058
1059         rcu_read_lock();
1060         bt = rcu_dereference(rq->q->blk_trace);
1061         if (likely(!bt)) {
1062                 rcu_read_unlock();
1063                 return;
1064         }
1065
1066         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0,
1067                                 BLK_TA_DRV_DATA, 0, len, data,
1068                                 blk_trace_request_get_cgid(rq));
1069         rcu_read_unlock();
1070 }
1071 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1072
1073 static void blk_register_tracepoints(void)
1074 {
1075         int ret;
1076
1077         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1078         WARN_ON(ret);
1079         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1080         WARN_ON(ret);
1081         ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1082         WARN_ON(ret);
1083         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1084         WARN_ON(ret);
1085         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1086         WARN_ON(ret);
1087         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1088         WARN_ON(ret);
1089         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1090         WARN_ON(ret);
1091         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1092         WARN_ON(ret);
1093         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1094         WARN_ON(ret);
1095         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1096         WARN_ON(ret);
1097         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1098         WARN_ON(ret);
1099         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1100         WARN_ON(ret);
1101         ret = register_trace_block_split(blk_add_trace_split, NULL);
1102         WARN_ON(ret);
1103         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1104         WARN_ON(ret);
1105         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1106         WARN_ON(ret);
1107 }
1108
1109 static void blk_unregister_tracepoints(void)
1110 {
1111         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1112         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1113         unregister_trace_block_split(blk_add_trace_split, NULL);
1114         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1115         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1116         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1117         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1118         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1119         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1120         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1121         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1122         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1123         unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1124         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1125         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1126
1127         tracepoint_synchronize_unregister();
1128 }
1129
1130 /*
1131  * struct blk_io_tracer formatting routines
1132  */
1133
1134 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1135 {
1136         int i = 0;
1137         int tc = t->action >> BLK_TC_SHIFT;
1138
1139         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1140                 rwbs[i++] = 'N';
1141                 goto out;
1142         }
1143
1144         if (tc & BLK_TC_FLUSH)
1145                 rwbs[i++] = 'F';
1146
1147         if (tc & BLK_TC_DISCARD)
1148                 rwbs[i++] = 'D';
1149         else if (tc & BLK_TC_WRITE)
1150                 rwbs[i++] = 'W';
1151         else if (t->bytes)
1152                 rwbs[i++] = 'R';
1153         else
1154                 rwbs[i++] = 'N';
1155
1156         if (tc & BLK_TC_FUA)
1157                 rwbs[i++] = 'F';
1158         if (tc & BLK_TC_AHEAD)
1159                 rwbs[i++] = 'A';
1160         if (tc & BLK_TC_SYNC)
1161                 rwbs[i++] = 'S';
1162         if (tc & BLK_TC_META)
1163                 rwbs[i++] = 'M';
1164 out:
1165         rwbs[i] = '\0';
1166 }
1167
1168 static inline
1169 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1170 {
1171         return (const struct blk_io_trace *)ent;
1172 }
1173
1174 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1175 {
1176         return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1177 }
1178
1179 static inline u64 t_cgid(const struct trace_entry *ent)
1180 {
1181         return *(u64 *)(te_blk_io_trace(ent) + 1);
1182 }
1183
1184 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1185 {
1186         return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1187 }
1188
1189 static inline u32 t_action(const struct trace_entry *ent)
1190 {
1191         return te_blk_io_trace(ent)->action;
1192 }
1193
1194 static inline u32 t_bytes(const struct trace_entry *ent)
1195 {
1196         return te_blk_io_trace(ent)->bytes;
1197 }
1198
1199 static inline u32 t_sec(const struct trace_entry *ent)
1200 {
1201         return te_blk_io_trace(ent)->bytes >> 9;
1202 }
1203
1204 static inline unsigned long long t_sector(const struct trace_entry *ent)
1205 {
1206         return te_blk_io_trace(ent)->sector;
1207 }
1208
1209 static inline __u16 t_error(const struct trace_entry *ent)
1210 {
1211         return te_blk_io_trace(ent)->error;
1212 }
1213
1214 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1215 {
1216         const __be64 *val = pdu_start(ent, has_cg);
1217         return be64_to_cpu(*val);
1218 }
1219
1220 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1221         bool has_cg);
1222
1223 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1224         bool has_cg)
1225 {
1226         char rwbs[RWBS_LEN];
1227         unsigned long long ts  = iter->ts;
1228         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1229         unsigned secs          = (unsigned long)ts;
1230         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1231
1232         fill_rwbs(rwbs, t);
1233
1234         trace_seq_printf(&iter->seq,
1235                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1236                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1237                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1238 }
1239
1240 static void blk_log_action(struct trace_iterator *iter, const char *act,
1241         bool has_cg)
1242 {
1243         char rwbs[RWBS_LEN];
1244         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1245
1246         fill_rwbs(rwbs, t);
1247         if (has_cg) {
1248                 u64 id = t_cgid(iter->ent);
1249
1250                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1251                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1252
1253                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1254                                 sizeof(blkcg_name_buf));
1255                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1256                                  MAJOR(t->device), MINOR(t->device),
1257                                  blkcg_name_buf, act, rwbs);
1258                 } else {
1259                         /*
1260                          * The cgid portion used to be "INO,GEN".  Userland
1261                          * builds a FILEID_INO32_GEN fid out of them and
1262                          * opens the cgroup using open_by_handle_at(2).
1263                          * While 32bit ino setups are still the same, 64bit
1264                          * ones now use the 64bit ino as the whole ID and
1265                          * no longer use generation.
1266                          *
1267                          * Regardless of the content, always output
1268                          * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1269                          * be mapped back to @id on both 64 and 32bit ino
1270                          * setups.  See __kernfs_fh_to_dentry().
1271                          */
1272                         trace_seq_printf(&iter->seq,
1273                                  "%3d,%-3d %llx,%-llx %2s %3s ",
1274                                  MAJOR(t->device), MINOR(t->device),
1275                                  id & U32_MAX, id >> 32, act, rwbs);
1276                 }
1277         } else
1278                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1279                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1280 }
1281
1282 static void blk_log_dump_pdu(struct trace_seq *s,
1283         const struct trace_entry *ent, bool has_cg)
1284 {
1285         const unsigned char *pdu_buf;
1286         int pdu_len;
1287         int i, end;
1288
1289         pdu_buf = pdu_start(ent, has_cg);
1290         pdu_len = pdu_real_len(ent, has_cg);
1291
1292         if (!pdu_len)
1293                 return;
1294
1295         /* find the last zero that needs to be printed */
1296         for (end = pdu_len - 1; end >= 0; end--)
1297                 if (pdu_buf[end])
1298                         break;
1299         end++;
1300
1301         trace_seq_putc(s, '(');
1302
1303         for (i = 0; i < pdu_len; i++) {
1304
1305                 trace_seq_printf(s, "%s%02x",
1306                                  i == 0 ? "" : " ", pdu_buf[i]);
1307
1308                 /*
1309                  * stop when the rest is just zeros and indicate so
1310                  * with a ".." appended
1311                  */
1312                 if (i == end && end != pdu_len - 1) {
1313                         trace_seq_puts(s, " ..) ");
1314                         return;
1315                 }
1316         }
1317
1318         trace_seq_puts(s, ") ");
1319 }
1320
1321 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1322 {
1323         char cmd[TASK_COMM_LEN];
1324
1325         trace_find_cmdline(ent->pid, cmd);
1326
1327         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1328                 trace_seq_printf(s, "%u ", t_bytes(ent));
1329                 blk_log_dump_pdu(s, ent, has_cg);
1330                 trace_seq_printf(s, "[%s]\n", cmd);
1331         } else {
1332                 if (t_sec(ent))
1333                         trace_seq_printf(s, "%llu + %u [%s]\n",
1334                                                 t_sector(ent), t_sec(ent), cmd);
1335                 else
1336                         trace_seq_printf(s, "[%s]\n", cmd);
1337         }
1338 }
1339
1340 static void blk_log_with_error(struct trace_seq *s,
1341                               const struct trace_entry *ent, bool has_cg)
1342 {
1343         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1344                 blk_log_dump_pdu(s, ent, has_cg);
1345                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1346         } else {
1347                 if (t_sec(ent))
1348                         trace_seq_printf(s, "%llu + %u [%d]\n",
1349                                          t_sector(ent),
1350                                          t_sec(ent), t_error(ent));
1351                 else
1352                         trace_seq_printf(s, "%llu [%d]\n",
1353                                          t_sector(ent), t_error(ent));
1354         }
1355 }
1356
1357 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1358 {
1359         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1360
1361         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1362                          t_sector(ent), t_sec(ent),
1363                          MAJOR(be32_to_cpu(__r->device_from)),
1364                          MINOR(be32_to_cpu(__r->device_from)),
1365                          be64_to_cpu(__r->sector_from));
1366 }
1367
1368 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1369 {
1370         char cmd[TASK_COMM_LEN];
1371
1372         trace_find_cmdline(ent->pid, cmd);
1373
1374         trace_seq_printf(s, "[%s]\n", cmd);
1375 }
1376
1377 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1378 {
1379         char cmd[TASK_COMM_LEN];
1380
1381         trace_find_cmdline(ent->pid, cmd);
1382
1383         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1384 }
1385
1386 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1387 {
1388         char cmd[TASK_COMM_LEN];
1389
1390         trace_find_cmdline(ent->pid, cmd);
1391
1392         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1393                          get_pdu_int(ent, has_cg), cmd);
1394 }
1395
1396 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1397                         bool has_cg)
1398 {
1399
1400         trace_seq_putmem(s, pdu_start(ent, has_cg),
1401                 pdu_real_len(ent, has_cg));
1402         trace_seq_putc(s, '\n');
1403 }
1404
1405 /*
1406  * struct tracer operations
1407  */
1408
1409 static void blk_tracer_print_header(struct seq_file *m)
1410 {
1411         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1412                 return;
1413         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1414                     "#  |     |     |           |   |   |\n");
1415 }
1416
1417 static void blk_tracer_start(struct trace_array *tr)
1418 {
1419         blk_tracer_enabled = true;
1420 }
1421
1422 static int blk_tracer_init(struct trace_array *tr)
1423 {
1424         blk_tr = tr;
1425         blk_tracer_start(tr);
1426         return 0;
1427 }
1428
1429 static void blk_tracer_stop(struct trace_array *tr)
1430 {
1431         blk_tracer_enabled = false;
1432 }
1433
1434 static void blk_tracer_reset(struct trace_array *tr)
1435 {
1436         blk_tracer_stop(tr);
1437 }
1438
1439 static const struct {
1440         const char *act[2];
1441         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1442                             bool has_cg);
1443 } what2act[] = {
1444         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1445         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1446         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1447         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1448         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1449         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1450         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1451         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1452         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1453         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1454         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1455         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1456         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1457         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1458 };
1459
1460 static enum print_line_t print_one_line(struct trace_iterator *iter,
1461                                         bool classic)
1462 {
1463         struct trace_array *tr = iter->tr;
1464         struct trace_seq *s = &iter->seq;
1465         const struct blk_io_trace *t;
1466         u16 what;
1467         bool long_act;
1468         blk_log_action_t *log_action;
1469         bool has_cg;
1470
1471         t          = te_blk_io_trace(iter->ent);
1472         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1473         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1474         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1475         has_cg     = t->action & __BLK_TA_CGROUP;
1476
1477         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1478                 log_action(iter, long_act ? "message" : "m", has_cg);
1479                 blk_log_msg(s, iter->ent, has_cg);
1480                 return trace_handle_return(s);
1481         }
1482
1483         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1484                 trace_seq_printf(s, "Unknown action %x\n", what);
1485         else {
1486                 log_action(iter, what2act[what].act[long_act], has_cg);
1487                 what2act[what].print(s, iter->ent, has_cg);
1488         }
1489
1490         return trace_handle_return(s);
1491 }
1492
1493 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1494                                                int flags, struct trace_event *event)
1495 {
1496         return print_one_line(iter, false);
1497 }
1498
1499 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1500 {
1501         struct trace_seq *s = &iter->seq;
1502         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1503         const int offset = offsetof(struct blk_io_trace, sector);
1504         struct blk_io_trace old = {
1505                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1506                 .time     = iter->ts,
1507         };
1508
1509         trace_seq_putmem(s, &old, offset);
1510         trace_seq_putmem(s, &t->sector,
1511                          sizeof(old) - offset + t->pdu_len);
1512 }
1513
1514 static enum print_line_t
1515 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1516                              struct trace_event *event)
1517 {
1518         blk_trace_synthesize_old_trace(iter);
1519
1520         return trace_handle_return(&iter->seq);
1521 }
1522
1523 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1524 {
1525         if ((iter->ent->type != TRACE_BLK) ||
1526             !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1527                 return TRACE_TYPE_UNHANDLED;
1528
1529         return print_one_line(iter, true);
1530 }
1531
1532 static int
1533 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1534 {
1535         /* don't output context-info for blk_classic output */
1536         if (bit == TRACE_BLK_OPT_CLASSIC) {
1537                 if (set)
1538                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1539                 else
1540                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1541         }
1542         return 0;
1543 }
1544
1545 static struct tracer blk_tracer __read_mostly = {
1546         .name           = "blk",
1547         .init           = blk_tracer_init,
1548         .reset          = blk_tracer_reset,
1549         .start          = blk_tracer_start,
1550         .stop           = blk_tracer_stop,
1551         .print_header   = blk_tracer_print_header,
1552         .print_line     = blk_tracer_print_line,
1553         .flags          = &blk_tracer_flags,
1554         .set_flag       = blk_tracer_set_flag,
1555 };
1556
1557 static struct trace_event_functions trace_blk_event_funcs = {
1558         .trace          = blk_trace_event_print,
1559         .binary         = blk_trace_event_print_binary,
1560 };
1561
1562 static struct trace_event trace_blk_event = {
1563         .type           = TRACE_BLK,
1564         .funcs          = &trace_blk_event_funcs,
1565 };
1566
1567 static int __init init_blk_tracer(void)
1568 {
1569         if (!register_trace_event(&trace_blk_event)) {
1570                 pr_warn("Warning: could not register block events\n");
1571                 return 1;
1572         }
1573
1574         if (register_tracer(&blk_tracer) != 0) {
1575                 pr_warn("Warning: could not register the block tracer\n");
1576                 unregister_trace_event(&trace_blk_event);
1577                 return 1;
1578         }
1579
1580         return 0;
1581 }
1582
1583 device_initcall(init_blk_tracer);
1584
1585 static int blk_trace_remove_queue(struct request_queue *q)
1586 {
1587         struct blk_trace *bt;
1588
1589         bt = rcu_replace_pointer(q->blk_trace, NULL,
1590                                  lockdep_is_held(&q->debugfs_mutex));
1591         if (bt == NULL)
1592                 return -EINVAL;
1593
1594         blk_trace_stop(bt);
1595
1596         put_probe_ref();
1597         synchronize_rcu();
1598         blk_trace_free(q, bt);
1599         return 0;
1600 }
1601
1602 /*
1603  * Setup everything required to start tracing
1604  */
1605 static int blk_trace_setup_queue(struct request_queue *q,
1606                                  struct block_device *bdev)
1607 {
1608         struct blk_trace *bt = NULL;
1609         int ret = -ENOMEM;
1610
1611         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1612         if (!bt)
1613                 return -ENOMEM;
1614
1615         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1616         if (!bt->msg_data)
1617                 goto free_bt;
1618
1619         bt->dev = bdev->bd_dev;
1620         bt->act_mask = (u16)-1;
1621
1622         blk_trace_setup_lba(bt, bdev);
1623
1624         rcu_assign_pointer(q->blk_trace, bt);
1625         get_probe_ref();
1626         return 0;
1627
1628 free_bt:
1629         blk_trace_free(q, bt);
1630         return ret;
1631 }
1632
1633 /*
1634  * sysfs interface to enable and configure tracing
1635  */
1636
1637 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1638                                          struct device_attribute *attr,
1639                                          char *buf);
1640 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1641                                           struct device_attribute *attr,
1642                                           const char *buf, size_t count);
1643 #define BLK_TRACE_DEVICE_ATTR(_name) \
1644         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1645                     sysfs_blk_trace_attr_show, \
1646                     sysfs_blk_trace_attr_store)
1647
1648 static BLK_TRACE_DEVICE_ATTR(enable);
1649 static BLK_TRACE_DEVICE_ATTR(act_mask);
1650 static BLK_TRACE_DEVICE_ATTR(pid);
1651 static BLK_TRACE_DEVICE_ATTR(start_lba);
1652 static BLK_TRACE_DEVICE_ATTR(end_lba);
1653
1654 static struct attribute *blk_trace_attrs[] = {
1655         &dev_attr_enable.attr,
1656         &dev_attr_act_mask.attr,
1657         &dev_attr_pid.attr,
1658         &dev_attr_start_lba.attr,
1659         &dev_attr_end_lba.attr,
1660         NULL
1661 };
1662
1663 struct attribute_group blk_trace_attr_group = {
1664         .name  = "trace",
1665         .attrs = blk_trace_attrs,
1666 };
1667
1668 static const struct {
1669         int mask;
1670         const char *str;
1671 } mask_maps[] = {
1672         { BLK_TC_READ,          "read"          },
1673         { BLK_TC_WRITE,         "write"         },
1674         { BLK_TC_FLUSH,         "flush"         },
1675         { BLK_TC_SYNC,          "sync"          },
1676         { BLK_TC_QUEUE,         "queue"         },
1677         { BLK_TC_REQUEUE,       "requeue"       },
1678         { BLK_TC_ISSUE,         "issue"         },
1679         { BLK_TC_COMPLETE,      "complete"      },
1680         { BLK_TC_FS,            "fs"            },
1681         { BLK_TC_PC,            "pc"            },
1682         { BLK_TC_NOTIFY,        "notify"        },
1683         { BLK_TC_AHEAD,         "ahead"         },
1684         { BLK_TC_META,          "meta"          },
1685         { BLK_TC_DISCARD,       "discard"       },
1686         { BLK_TC_DRV_DATA,      "drv_data"      },
1687         { BLK_TC_FUA,           "fua"           },
1688 };
1689
1690 static int blk_trace_str2mask(const char *str)
1691 {
1692         int i;
1693         int mask = 0;
1694         char *buf, *s, *token;
1695
1696         buf = kstrdup(str, GFP_KERNEL);
1697         if (buf == NULL)
1698                 return -ENOMEM;
1699         s = strstrip(buf);
1700
1701         while (1) {
1702                 token = strsep(&s, ",");
1703                 if (token == NULL)
1704                         break;
1705
1706                 if (*token == '\0')
1707                         continue;
1708
1709                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1710                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1711                                 mask |= mask_maps[i].mask;
1712                                 break;
1713                         }
1714                 }
1715                 if (i == ARRAY_SIZE(mask_maps)) {
1716                         mask = -EINVAL;
1717                         break;
1718                 }
1719         }
1720         kfree(buf);
1721
1722         return mask;
1723 }
1724
1725 static ssize_t blk_trace_mask2str(char *buf, int mask)
1726 {
1727         int i;
1728         char *p = buf;
1729
1730         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1731                 if (mask & mask_maps[i].mask) {
1732                         p += sprintf(p, "%s%s",
1733                                     (p == buf) ? "" : ",", mask_maps[i].str);
1734                 }
1735         }
1736         *p++ = '\n';
1737
1738         return p - buf;
1739 }
1740
1741 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1742                                          struct device_attribute *attr,
1743                                          char *buf)
1744 {
1745         struct block_device *bdev = dev_to_bdev(dev);
1746         struct request_queue *q = bdev_get_queue(bdev);
1747         struct blk_trace *bt;
1748         ssize_t ret = -ENXIO;
1749
1750         mutex_lock(&q->debugfs_mutex);
1751
1752         bt = rcu_dereference_protected(q->blk_trace,
1753                                        lockdep_is_held(&q->debugfs_mutex));
1754         if (attr == &dev_attr_enable) {
1755                 ret = sprintf(buf, "%u\n", !!bt);
1756                 goto out_unlock_bdev;
1757         }
1758
1759         if (bt == NULL)
1760                 ret = sprintf(buf, "disabled\n");
1761         else if (attr == &dev_attr_act_mask)
1762                 ret = blk_trace_mask2str(buf, bt->act_mask);
1763         else if (attr == &dev_attr_pid)
1764                 ret = sprintf(buf, "%u\n", bt->pid);
1765         else if (attr == &dev_attr_start_lba)
1766                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1767         else if (attr == &dev_attr_end_lba)
1768                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1769
1770 out_unlock_bdev:
1771         mutex_unlock(&q->debugfs_mutex);
1772         return ret;
1773 }
1774
1775 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1776                                           struct device_attribute *attr,
1777                                           const char *buf, size_t count)
1778 {
1779         struct block_device *bdev = dev_to_bdev(dev);
1780         struct request_queue *q = bdev_get_queue(bdev);
1781         struct blk_trace *bt;
1782         u64 value;
1783         ssize_t ret = -EINVAL;
1784
1785         if (count == 0)
1786                 goto out;
1787
1788         if (attr == &dev_attr_act_mask) {
1789                 if (kstrtoull(buf, 0, &value)) {
1790                         /* Assume it is a list of trace category names */
1791                         ret = blk_trace_str2mask(buf);
1792                         if (ret < 0)
1793                                 goto out;
1794                         value = ret;
1795                 }
1796         } else {
1797                 if (kstrtoull(buf, 0, &value))
1798                         goto out;
1799         }
1800
1801         mutex_lock(&q->debugfs_mutex);
1802
1803         bt = rcu_dereference_protected(q->blk_trace,
1804                                        lockdep_is_held(&q->debugfs_mutex));
1805         if (attr == &dev_attr_enable) {
1806                 if (!!value == !!bt) {
1807                         ret = 0;
1808                         goto out_unlock_bdev;
1809                 }
1810                 if (value)
1811                         ret = blk_trace_setup_queue(q, bdev);
1812                 else
1813                         ret = blk_trace_remove_queue(q);
1814                 goto out_unlock_bdev;
1815         }
1816
1817         ret = 0;
1818         if (bt == NULL) {
1819                 ret = blk_trace_setup_queue(q, bdev);
1820                 bt = rcu_dereference_protected(q->blk_trace,
1821                                 lockdep_is_held(&q->debugfs_mutex));
1822         }
1823
1824         if (ret == 0) {
1825                 if (attr == &dev_attr_act_mask)
1826                         bt->act_mask = value;
1827                 else if (attr == &dev_attr_pid)
1828                         bt->pid = value;
1829                 else if (attr == &dev_attr_start_lba)
1830                         bt->start_lba = value;
1831                 else if (attr == &dev_attr_end_lba)
1832                         bt->end_lba = value;
1833         }
1834
1835 out_unlock_bdev:
1836         mutex_unlock(&q->debugfs_mutex);
1837 out:
1838         return ret ? ret : count;
1839 }
1840 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1841
1842 #ifdef CONFIG_EVENT_TRACING
1843
1844 /**
1845  * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string.
1846  * @rwbs:       buffer to be filled
1847  * @opf:        request operation type (REQ_OP_XXX) and flags for the tracepoint
1848  *
1849  * Description:
1850  *     Maps each request operation and flag to a single character and fills the
1851  *     buffer provided by the caller with resulting string.
1852  *
1853  **/
1854 void blk_fill_rwbs(char *rwbs, blk_opf_t opf)
1855 {
1856         int i = 0;
1857
1858         if (opf & REQ_PREFLUSH)
1859                 rwbs[i++] = 'F';
1860
1861         switch (opf & REQ_OP_MASK) {
1862         case REQ_OP_WRITE:
1863                 rwbs[i++] = 'W';
1864                 break;
1865         case REQ_OP_DISCARD:
1866                 rwbs[i++] = 'D';
1867                 break;
1868         case REQ_OP_SECURE_ERASE:
1869                 rwbs[i++] = 'D';
1870                 rwbs[i++] = 'E';
1871                 break;
1872         case REQ_OP_FLUSH:
1873                 rwbs[i++] = 'F';
1874                 break;
1875         case REQ_OP_READ:
1876                 rwbs[i++] = 'R';
1877                 break;
1878         default:
1879                 rwbs[i++] = 'N';
1880         }
1881
1882         if (opf & REQ_FUA)
1883                 rwbs[i++] = 'F';
1884         if (opf & REQ_RAHEAD)
1885                 rwbs[i++] = 'A';
1886         if (opf & REQ_SYNC)
1887                 rwbs[i++] = 'S';
1888         if (opf & REQ_META)
1889                 rwbs[i++] = 'M';
1890         if (opf & REQ_ATOMIC)
1891                 rwbs[i++] = 'U';
1892
1893         rwbs[i] = '\0';
1894 }
1895 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1896
1897 #endif /* CONFIG_EVENT_TRACING */
1898